4 * Copyright (C) 2007 Marc Hoffman <marc.hoffman@analog.com>
5 * Copyright (c) 2006 Michael Benjamin <michael.benjamin@analog.com>
7 * This file is part of FFmpeg.
9 * FFmpeg is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public
11 * License as published by the Free Software Foundation; either
12 * version 2.1 of the License, or (at your option) any later version.
14 * FFmpeg is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with FFmpeg; if not, write to the Free Software
21 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
24 #include "libavcodec/avcodec.h"
25 #include "libavcodec/dsputil.h"
26 #include "dsputil_bfin.h"
30 static void bfin_idct_add (uint8_t *dest, int line_size, DCTELEM *block)
33 ff_bfin_add_pixels_clamped (block, dest, line_size);
36 static void bfin_idct_put (uint8_t *dest, int line_size, DCTELEM *block)
39 ff_bfin_put_pixels_clamped (block, dest, line_size);
43 static void bfin_clear_blocks (DCTELEM *blocks)
45 // This is just a simple memset.
50 "LSETUP(clear_blocks_blkfn_lab,clear_blocks_blkfn_lab)LC0=P0;"
51 "clear_blocks_blkfn_lab:"
53 ::"a" (blocks):"P0","I0","R0");
58 static void bfin_put_pixels8 (uint8_t *block, const uint8_t *pixels, int line_size, int h)
60 ff_bfin_put_pixels8uc (block, pixels, pixels, line_size, line_size, h);
63 static void bfin_put_pixels8_x2(uint8_t *block, const uint8_t *pixels, int line_size, int h)
65 ff_bfin_put_pixels8uc (block, pixels, pixels+1, line_size, line_size, h);
68 static void bfin_put_pixels8_y2 (uint8_t *block, const uint8_t *pixels, int line_size, int h)
70 ff_bfin_put_pixels8uc (block, pixels, pixels+line_size, line_size, line_size, h);
73 static void bfin_put_pixels8_xy2 (uint8_t *block, const uint8_t *s0, int line_size, int h)
75 ff_bfin_z_put_pixels8_xy2 (block,s0,line_size, line_size, h);
78 static void bfin_put_pixels16 (uint8_t *block, const uint8_t *pixels, int line_size, int h)
80 ff_bfin_put_pixels16uc (block, pixels, pixels, line_size, line_size, h);
83 static void bfin_put_pixels16_x2 (uint8_t *block, const uint8_t *pixels, int line_size, int h)
85 ff_bfin_put_pixels16uc (block, pixels, pixels+1, line_size, line_size, h);
88 static void bfin_put_pixels16_y2 (uint8_t *block, const uint8_t *pixels, int line_size, int h)
90 ff_bfin_put_pixels16uc (block, pixels, pixels+line_size, line_size, line_size, h);
93 static void bfin_put_pixels16_xy2 (uint8_t *block, const uint8_t *s0, int line_size, int h)
95 ff_bfin_z_put_pixels16_xy2 (block,s0,line_size, line_size, h);
98 static void bfin_put_pixels8_nornd (uint8_t *block, const uint8_t *pixels, int line_size, int h)
100 ff_bfin_put_pixels8uc_nornd (block, pixels, pixels, line_size, h);
103 static void bfin_put_pixels8_x2_nornd (uint8_t *block, const uint8_t *pixels, int line_size, int h)
105 ff_bfin_put_pixels8uc_nornd (block, pixels, pixels+1, line_size, h);
108 static void bfin_put_pixels8_y2_nornd (uint8_t *block, const uint8_t *pixels, int line_size, int h)
110 ff_bfin_put_pixels8uc_nornd (block, pixels, pixels+line_size, line_size, h);
114 static void bfin_put_pixels16_nornd (uint8_t *block, const uint8_t *pixels, int line_size, int h)
116 ff_bfin_put_pixels16uc_nornd (block, pixels, pixels, line_size, h);
119 static void bfin_put_pixels16_x2_nornd (uint8_t *block, const uint8_t *pixels, int line_size, int h)
121 ff_bfin_put_pixels16uc_nornd (block, pixels, pixels+1, line_size, h);
124 static void bfin_put_pixels16_y2_nornd (uint8_t *block, const uint8_t *pixels, int line_size, int h)
126 ff_bfin_put_pixels16uc_nornd (block, pixels, pixels+line_size, line_size, h);
129 static int bfin_pix_abs16 (void *c, uint8_t *blk1, uint8_t *blk2, int line_size, int h)
131 return ff_bfin_z_sad16x16 (blk1,blk2,line_size,line_size,h);
134 static int bfin_vsad_intra16 (void *c, uint8_t *blk1, uint8_t *dummy, int stride, int h) {
135 return ff_bfin_z_sad16x16 (blk1,blk1+stride,stride<<1,stride<<1,h);
138 static int bfin_vsad (void *c, uint8_t *blk1, uint8_t *blk2, int stride, int h) {
139 return ff_bfin_z_sad16x16 (blk1,blk1+stride,stride<<1,stride<<1,h)
140 + ff_bfin_z_sad16x16 (blk2,blk2+stride,stride<<1,stride<<1,h);
143 static uint8_t vtmp_blk[256] attribute_l1_data_b;
145 static int bfin_pix_abs16_x2 (void *c, uint8_t *blk1, uint8_t *blk2, int line_size, int h)
147 ff_bfin_put_pixels16uc (vtmp_blk, blk2, blk2+1, 16, line_size, h);
148 return ff_bfin_z_sad16x16 (blk1, vtmp_blk, line_size, 16, h);
151 static int bfin_pix_abs16_y2 (void *c, uint8_t *blk1, uint8_t *blk2, int line_size, int h)
153 ff_bfin_put_pixels16uc (vtmp_blk, blk2, blk2+line_size, 16, line_size, h);
154 return ff_bfin_z_sad16x16 (blk1, vtmp_blk, line_size, 16, h);
157 static int bfin_pix_abs16_xy2 (void *c, uint8_t *blk1, uint8_t *blk2, int line_size, int h)
159 ff_bfin_z_put_pixels16_xy2 (vtmp_blk, blk2, 16, line_size, h);
160 return ff_bfin_z_sad16x16 (blk1, vtmp_blk, line_size, 16, h);
163 static int bfin_pix_abs8 (void *c, uint8_t *blk1, uint8_t *blk2, int line_size, int h)
165 return ff_bfin_z_sad8x8 (blk1,blk2,line_size,line_size, h);
168 static int bfin_pix_abs8_x2 (void *c, uint8_t *blk1, uint8_t *blk2, int line_size, int h)
170 ff_bfin_put_pixels8uc (vtmp_blk, blk2, blk2+1, 8, line_size, h);
171 return ff_bfin_z_sad8x8 (blk1, vtmp_blk, line_size, 8, h);
174 static int bfin_pix_abs8_y2 (void *c, uint8_t *blk1, uint8_t *blk2, int line_size, int h)
176 ff_bfin_put_pixels8uc (vtmp_blk, blk2, blk2+line_size, 8, line_size, h);
177 return ff_bfin_z_sad8x8 (blk1, vtmp_blk, line_size, 8, h);
180 static int bfin_pix_abs8_xy2 (void *c, uint8_t *blk1, uint8_t *blk2, int line_size, int h)
182 ff_bfin_z_put_pixels8_xy2 (vtmp_blk, blk2, 8, line_size, h);
183 return ff_bfin_z_sad8x8 (blk1, vtmp_blk, line_size, 8, h);
189 start on 2/11 100 frames of 352x240@25 compiled with no optimization -g debugging
191 6.360s ~ 1.58x off with -O2
192 5.740s ~ 1.43x off with idcts
194 2.64s 2/20 same sman.mp4 decode only
198 void ff_dsputil_init_bfin( DSPContext* c, AVCodecContext *avctx )
200 const int high_bit_depth = avctx->bits_per_raw_sample > 8;
202 c->diff_pixels = ff_bfin_diff_pixels;
203 c->put_pixels_clamped = ff_bfin_put_pixels_clamped;
204 c->add_pixels_clamped = ff_bfin_add_pixels_clamped;
207 c->get_pixels = ff_bfin_get_pixels;
208 c->clear_blocks = bfin_clear_blocks;
209 c->pix_sum = ff_bfin_pix_sum;
210 c->pix_norm1 = ff_bfin_pix_norm1;
212 c->sad[0] = bfin_pix_abs16;
213 c->sad[1] = bfin_pix_abs8;
215 /* c->vsad[0] = bfin_vsad; */
216 /* c->vsad[4] = bfin_vsad_intra16; */
218 /* TODO [0] 16 [1] 8 */
219 c->pix_abs[0][0] = bfin_pix_abs16;
220 c->pix_abs[0][1] = bfin_pix_abs16_x2;
221 c->pix_abs[0][2] = bfin_pix_abs16_y2;
222 c->pix_abs[0][3] = bfin_pix_abs16_xy2;
224 c->pix_abs[1][0] = bfin_pix_abs8;
225 c->pix_abs[1][1] = bfin_pix_abs8_x2;
226 c->pix_abs[1][2] = bfin_pix_abs8_y2;
227 c->pix_abs[1][3] = bfin_pix_abs8_xy2;
230 c->sse[0] = ff_bfin_sse16;
231 c->sse[1] = ff_bfin_sse8;
232 c->sse[2] = ff_bfin_sse4;
234 if (!high_bit_depth) {
235 c->put_pixels_tab[0][0] = bfin_put_pixels16;
236 c->put_pixels_tab[0][1] = bfin_put_pixels16_x2;
237 c->put_pixels_tab[0][2] = bfin_put_pixels16_y2;
238 c->put_pixels_tab[0][3] = bfin_put_pixels16_xy2;
240 c->put_pixels_tab[1][0] = bfin_put_pixels8;
241 c->put_pixels_tab[1][1] = bfin_put_pixels8_x2;
242 c->put_pixels_tab[1][2] = bfin_put_pixels8_y2;
243 c->put_pixels_tab[1][3] = bfin_put_pixels8_xy2;
245 c->put_no_rnd_pixels_tab[1][0] = bfin_put_pixels8_nornd;
246 c->put_no_rnd_pixels_tab[1][1] = bfin_put_pixels8_x2_nornd;
247 c->put_no_rnd_pixels_tab[1][2] = bfin_put_pixels8_y2_nornd;
248 /* c->put_no_rnd_pixels_tab[1][3] = ff_bfin_put_pixels8_xy2_nornd; */
250 c->put_no_rnd_pixels_tab[0][0] = bfin_put_pixels16_nornd;
251 c->put_no_rnd_pixels_tab[0][1] = bfin_put_pixels16_x2_nornd;
252 c->put_no_rnd_pixels_tab[0][2] = bfin_put_pixels16_y2_nornd;
253 /* c->put_no_rnd_pixels_tab[0][3] = ff_bfin_put_pixels16_xy2_nornd; */
256 if (avctx->bits_per_raw_sample <= 8) {
257 if (avctx->dct_algo == FF_DCT_AUTO)
258 c->fdct = ff_bfin_fdct;
260 if (avctx->idct_algo == FF_IDCT_VP3) {
261 c->idct_permutation_type = FF_NO_IDCT_PERM;
262 c->idct = ff_bfin_vp3_idct;
263 c->idct_add = ff_bfin_vp3_idct_add;
264 c->idct_put = ff_bfin_vp3_idct_put;
265 } else if (avctx->idct_algo == FF_IDCT_AUTO) {
266 c->idct_permutation_type = FF_NO_IDCT_PERM;
267 c->idct = ff_bfin_idct;
268 c->idct_add = bfin_idct_add;
269 c->idct_put = bfin_idct_put;